Cohort Characteristics

The retrospective pan-solid-tumor evaluable cohort (N = 86) represented patients with non-small cell lung cancer (31.4%, N = 27), breast cancer (14.0%, N = 12), small cell lung cancer (18.6%, N = 16), and 11 additional cancer types (36.0%, N = 31). Median age of the cohort was 65, and 50.0% (N = 43) of the patients were female. Twenty-three percent of the cohort (N = 20) were Asian, Black, African American, or a race other than white. The majority of patients had stage IV disease (88.4%, N = 76). Thirty-eight percent of the cohort (N = 33) were treated with ICI therapy alone and 61.6% (N = 53) were treated with ICI + chemotherapy combinations (Table 1). The majority of patients in the cohort (53.5%, N = 46) were treated with an ICI agent as first-line therapy, while 45.3% (N = 39) of the patients were treated with ICI at or beyond second line (one patient was unknown); further stratified by cancer type in Supplementary Materials Table S1. The median time from baseline testing to ICI start was 15.0 days and the median time from ICI start to on-treatment testing was 91.0 days. Timing of on-treatment testing, dates and outcomes of rw-imaging, and death events when applicable in each group are shown in Fig. 1b.

Table 1 Clinical characteristics of patients evaluable for molecular response analysisctDNA TF Correlation with Tumor-Informed TF Outperforms Mean VAF

The calculation of the ctDNA TF is described in “Methods” (see Fig. 2). TF linearity was evaluated in a historical pan-cancer cohort (N = 1659 samples) using the tumor-informed TF as a reference (see “Methods”). We compared the linearity of ctDNA TF to that of mean VAF, a common estimate for TF in liquid biopsy. The linearity of ctDNA TF (Pearson R = 0.9, Spearman R = 0.8, slope = 0.9, P < 0.0001) outperformed the mean VAF (Pearson R = 0.7, Spearman R = 0.6, slope = 0.6, P < 0.0001, Fig. 3).

Fig. 3

Comparison of ctDNA TF with a tumor-informed ctDNA TF. ctDNA circulating tumor DNA, TF tumor fraction, VAF variant allele frequency

We also compared ctDNA TF and mean VAF using randomly selected liquid biopsy samples that underwent LPWGS and ichorCNA TF estimates (see “Methods”, N = 375), an orthogonal measure of ctDNA that had correlation with the tumor-informed ctDNA TF (N = 79 clinical tumor isolates that underwent LPWGS, Pearson R2 = 0.844, Spearman R2 = 0.771, slope = 0.884) [24]. The linearity of the ctDNA TF (Pearson R = 0.9, Spearman R = 0.7, slope = 0.7, P < 0.0001) outperformed the mean VAF (Pearson R = 0.3, Spearman R = 0.4, slope = 0.2, P < 0.0001, Supplementary Materials Fig. S1).

MR Association with RW-Outcomes

We first evaluated whether MR was associated with rwOS and rwPFS in the evaluable pan-solid tumor cohort (N = 86, ICI ± chemotherapy). Median time from ICI start to on-treatment testing was 88.0 days for patients with MR and 91.5 days for patients with nMR. rwOS was significantly longer among patients with MR (N = 42) versus patients with nMR (N = 44); median rwOS was 18.2 months for patients with MR and 11.3 months for patients with nMR (HR 0.4, 95% CI 0.2–0.7, P = 0.004, Fig. 4a). rwPFS was significantly longer among patients with MR versus patients with nMR, median rwPFS was 8.7 months for patients with MR and 4.0 months for patients with nMR, (HR 0.4, 95% CI 0.2–0.8, P = 0.005, Fig. 4b).

Fig. 4

Association of ctDNA MR in patients treated with any ICI ± chemotherapy regimen and a rwOS (HR 0.4, P = 0.004) and b rwPFS (HR 0.4, P = 0.005). Association of ctDNA MR in patients treated with ICI monotherapy and c rwOS (HR 0.2, P = 0.02) and d rwPFS (HR 0.2, P = 0.01). ctDNA circulating tumor DNA, HR hazard ratio, ICI immune checkpoint inhibitor, MR molecular response, rwOS real-world overall survival, rwPFS real-world progression free survival

Next we evaluated the association of MR with rwOS and rwPFS in the ICI monotherapy cohort (N = 34) and in the ICI + chemotherapy cohort separately (N = 53). In the ICI monotherapy cohort, rwOS was significantly longer among patients with MR versus patients with nMR, median rwOS was not reached for patients with MR versus 11.3 months for patients with nMR, (HR 0.2, 95% CI 0.05–0.7, P = 0.02, Fig. 4c). In this same ICI alone cohort, rwPFS was also significantly longer in the patients with MR relative to the patients with nMR (median rwPFS was 17.9 months for patients with MR versus 4.8 months for patients with nMR; HR 0.2, 95% CI 0.08–0.7, P = 0.01, Fig. 4d).

Similar trends were seen in rwOS and rwPFS within the ICI + CT cohort, though statistical significance was not reached (Supplementary Materials Fig. S2). rwOS was longer among patients with MR versus those with nMR, with a median rwOS of 14.4 months for patients with MR versus 10.0 months for patients with nMR (HR 0.5, 95% CI 0.2–1.1, P = 0.07, Fig. S2a). rwPFS was longer in the patients with MR relative to the patients with nMR (median rwPFS was 4.8 months for patients with MR versus 3.3 months for patients with nMR; HR 0.5, 95% CI 0.2–1.3, P = 0.2, Fig. S2b).

Given the heterogeneity of cancer types within our pan-cancer cohort, we reevaluated the association of MR with rw-outcomes, while accounting for cancer type. MR was significantly associated with rwOS after adjustment for cancer type (HR 0.4, 95% CI 0.2–0.7, P = 0.004, Supplementary Materials Table S2) and was significantly associated with rwPFS after adjustment for cancer type (HR 0.2, CI 0.1–0.5, P < 0.001, Supplementary Materials Table S3).

Nine patients (9%) whose baseline and on-treatment TF were both below the LOB were considered clinically as TF undetected. Timing of on-treatment testing, rw-imaging, and death events when applicable for patients consistently below LOB are shown in Supplementary Materials Fig. S3. Patient demographics for patients consistently below LOB are shown in Supplementary Materials Table S4. These patients had no death events and were not evaluable for rwOS analysis.

Predictive Value of MR in Addition to RW-Imaging

In the rw-imaging subcohort (N = 51, see “Methods”), we evaluated whether MR had predictive value beyond rw-imaging. Patient demographics for this cohort are shown in Supplementary Materials Table S5. Additionally, timing of on-treatment testing, rw-imaging, and death events are shown in Supplementary Materials Fig. S5. While the majority of patients within this cohort had concordant MR and rw-imaging response results, a subset (35.3%, N = 18) had discordant results (Fig. 5a).

Fig. 5

a Risk stratifications based on MR, rw-imaging response and b predicted rwOS probability using a model combining MR and rw-imaging response (see “Methods”). CR complete response, MR molecular response, PD progressive disease, PR partial response, rwOS real-world overall survival, SD stable disease

We compared a reduced model of rw-imaging only versus a full model (rw-imaging combined with MR) to investigate whether the addition of MR to rw-imaging could better predict rwOS than rw-imaging alone (see “Methods”). Our results show that the full model (HR 0.3, CI 0.1–0.8, P = 0.02 for MR vs. nMR, HR 0.7, CI 0.3–1.5, P = 0.3 for rw-imaging response vs. rw-imaging non-response) is superior in predicting rwOS (P = 0.02) than the reduced model of response via rw-imaging alone (HR 0.5, CI 0.2–1.2, P = 0.1, for rw-imaging response vs. rw-imaging non-response). Patients with nMR and rw-imaging response (N = 13) had a shorter model-predicted median rwOS than patients with MR and rw-imaging response (N = 19, 9.9 months vs. 16.0 months, respectively, Fig. 5b). Similarly, patients with nMR and PD via rw-imaging (N = 14) had a shorter model-predicted median rwOS than patients with MR and PD (N = 5, 7.8 months vs. 15.3 months, respectively, Fig. 5b).

MR Association with RW-Outcomes in Patients with Early On-treatment Timepoints

Given that standard of care CT imaging typically occurs 12 weeks post ICI treatment initiation, we evaluated the association of MR with rwOS and rwPFS for the subset of patients with an on-treatment timepoint ≤ 12 weeks of ICI treatment initiation (N = 40). Timing of on-treatment testing, rw-imaging, and death events are shown in Supplementary Materials Fig. S6. rwOS was significantly longer for patients with MR versus patients with nMR, with median rwOS of 22.4 months for patients with MR versus 11.4 months for patients with nMR (HR 0.2, 95% CI 0.08–0.7, P = 0.01, Supplementary Materials Fig. S7). Similarly, rwPFS was longer among patients with MR versus patients with nMR, though this finding was not statistically significant, with median rwPFS of 8.7 months for patients with MR versus 4.8 months for patients with nMR (HR 0.5, 95% CI 0.2–1.2, P = 0.1 Supplementary Materials Fig. S7). For five patients that had nMR and PD via rw-imaging following their on-treatment liquid biopsy, the median lead time from molecular progression to PD via rw-imaging was 134.0 days.

Sensitivity Analyses

Patients with MR showed consistently longer rwOS and rwPFS than patients with nMR across a broad range of potential MR thresholds (0–90% reduction in TF), demonstrating that our results are robust to the choice of percent-reduction threshold in TF. For rwOS, both the HRs and the corresponding p values trended downward as the threshold for MR became more stringent (Supplementary Materials Fig. S8). For rwPFS, the HRs and p values were similar across the explored range of MR thresholds (Supplementary Materials Fig. S8).

To assess how soon after ICI initiation we could detect significant separation in outcomes between patients with MR and patients with nMR, we reran our analysis using subcohorts with incrementally smaller on-treatment testing windows, from 182 to 28 days post ICI initiation. We observed significant separation in rwOS down to 42 days post ICI (N = 23) and significant separation in rwPFS down to 112 days post ICI (N = 61) (Supplementary Materials Fig. S9).